Machine Learning-Based Radio Environment Maps for 4G/5G Networks

The exponential increase in wireless device usage, combined with a burgeoning demand for bandwidth-intensive applications, underscores the need for significant enhancements in network infrastructure, particularly within the realms of fifth-generation (5G) and beyond 5G (B5G) technologies. These enhancements must effectively address comprehensive network coverage, minimize latency, and improve overall efficiency. Consequently, this requires the deployment of advanced methodologies aimed at optimizing network performance, which poses complex challenges for network managers and operators. The adoption of machine learning (ML) techniques offers a formidable framework for predictive analytics in this domain, to effectively and efficiently improve network performance. This research paper explores the application of various boosting ML methodologies, with a focus on the creation of radio environmental maps (REMs). We use a new boosting method, namely the Gaussian Process Boosting (GPBoost) that has not yet been applied to an electromagnetics problem. Moreover, we solve a multi-output problem and present a model that can be used to predict both reference signal received power (RSRP) and reference signal received quality (RSRQ). This analytical approach enables accurate forecasting of critical network metrics that are crucial to evaluate network quality and operational efficiency. This study uses measurement data from the literature conducted on both 4G and 5G networks. The initial findings indicate that these methodologies considerably improve the accuracy of the predicted network performance metrics.